Split-likelihood methodology

What the score is — and what it isn't

The split-likelihood score is a transparent, rule-based 0–100 signal that measures how structurally primed a stock is to declare a forward split — how closely it resembles companies that have split in the past. It is a heuristic, not advice, and never a probability or a prediction. A split is a private board decision; no public signal can forecast it. A high score describes a setup, nothing more.

The score is gated to actively trading US common equity priced at $5 and up. ETFs, ADRs, and non-USD listings are excluded; a stock below the price floor is flagged low-price and not numerically scored. Reverse splits — a different, distress-driven phenomenon — are out of scope.

Provenance. Price curve back-tested against 2012+ split-vs-population lift; the four-factor signal is validated by a matched-control back-test (AUC 0.64, High band 2.4× base rate); factor weights are expert-set (validated, not reweighted) and timing is expert-set.

The four factors and their weights

Four additive factors are each scored on a 0–1 scale from public data, then combined by the weights below. When a factor's data is missing it is dropped and the remaining weights are renormalized, so the available factors always sum to 100%.

The four additive split-likelihood factors and their default weights
Factor	Weight	What it measures
Absolute price	40%	How high the raw share price is — the optical level boards actually react to.
Personal split threshold	28%	Price relative to the company's own historical pre-split prices. A repeat splitter back near the level it split at before scores high.
Split track record	20%	How many times the company has split before — splitting is a habit.
Proximity to 52-week high	12%	Price as a share of the 52-week high — boards rarely split a falling stock.

The factor weights are expert-set (back-test-validated, not reweighted): a weight search on the back-test sample improved separation only marginally and only via degenerate vectors, so the interpretable expert weights are kept.

The timing multiplier

Recency lives in exactly one place. After the four factors are combined, a single non-additive multiplier scales the result by how long ago the company last split — companies rarely split twice in quick succession, and a long-stale split habit counts for less. Timing is expert-set.

The single non-additive timing multiplier applied to the weighted factor sum
Situation	Multiplier
Recently split — cooldown	0.40×
Split history in the active window	1.00×
Last split 10–20 years ago	0.85×
Last split over 20 years ago	0.70×
No prior split on record	0.60×

Band cutoffs

The final 0–100 score maps to one of four bands. The ranking page surfaces Watch-band-and-up candidates.

Score-to-band cutoffs
Band	Score range
Low	0–24
Watch	25–49
Elevated	50–74
High	75–100

The calibrated price curve and the data behind it

Split behaviour is non-stationary: 1990s split mania clustered at $50–100, whereas modern up-market splits are rarer and happen at much higher prices. So the price curve is calibrated on modern-regime data only (splits on or after 2012). Across that window the median pre-split price is roughly $85, and about 86% of splitters had split before — splitting really is a repeat habit.

The price factor is the lift profile: for each price band we divide the band's share of meaningful forward splits by its share of the eligible universe. Lift above 1.0 means splits are over-represented there. The shape is bimodal — a classic-split hump at $50–100, a mid-price trough, and a dominant peak at $500–4,000 (modern mega-cap splits). The high-price tail is expert-set, since almost no companies live there.

Split-vs-population lift by price band, with the live calibrated price sub-score
Price band	Universe	Splits	Lift	Sub-score	Note
$5–25	61	3	0.16×	8%	Thin (n=3); well below baseline. Sub-score expert-floored.
$25–50	99	15	0.49×	18%	Below baseline — sub-$50 splits are rare once stock dividends are removed.
$50–100	162	76	1.52×	57%	Classic-split hump (n=76).
$100–250	223	66	0.96×	36%	~baseline (lift ≈ 1.0).
$250–500	111	20	0.59×	22%	Mid-price trough floor.
$500–1,000	25	18	2.34×	88%	Second hump (n=18).
$1,000–4,000	17	14	2.68×	100%	Peak — modern up-market mega-cap splits (n=14).
$4,000–8,000	1	1	3.25×	55%	n=1 — not calibrated; expert-set tail.
$8,000–15,000	0	2	—	20%	No universe baseline; expert-set tail.
$15,000+	0	0	—	12%	No data; expert-set tail (BRK.A-type, rarely splits).

How well the signal works — precision, recall & lift

We report precision, recall, and lift only — never accuracy. Forward splits are an extreme-minority event (≈15 modern splits a year across a ~700-company universe), so a model that says "never splits" scores ~99% accuracy while being useless. Precision, recall, and lift are the only honest metrics.

In a matched-control back-test (214 future splitters vs. 926 matched non-splitters, each scored 12 months before the event), the live signal separates the two groups with AUC 0.64 (0.5 is random) and a clean monotonic lift across bands — the validation that the score carries real signal.

Per-band back-test performance: share of future splitters and lift over the sample base rate
Band	Score range	Share of splitters	Lift
Low	0–24	22%	0.70×
Watch	25–49	25%	0.75×
Elevated	50–74	41%	1.32×
High	75–100	13%	2.36×

Read by threshold ("this band and up"), recall is the share of all splitters captured and the population-adjusted precision is the real-world rate after correcting for the matched sample's inflated base rate:

Threshold precision, recall and lift from the matched-control back-test
Threshold	Recall	Precision (population-adjusted)	Lift
Watch and up (≥25)	79%	2.3%	1.13×
Elevated and up (≥50)	54%	3.3%	1.47×
High and up (≥75)	13%	6.6%	2.36×

Known limits and disclaimers

Not a prediction. A split is a discretionary board decision. The score measures structural resemblance to past splitters, not the chance a split is coming — a strong candidate can sit unsplit for years (Berkshire Hathaway never split its A shares).
Forward splits only. Reverse splits are excluded; they are a different, distress-driven phenomenon.
Modest separation. An AUC of 0.64 is real but modest. The control pool is itself made of mature repeat-splitters, so track-record and proximity discriminate less here than they would across the full, younger ranking universe.
Sparse high-price tail. Above $4,000 there are almost no companies to calibrate against, so that part of the price curve is expert-set, not back-tested.
Not investment advice. The signal is an educational, informational tool — nothing here is a recommendation to buy, sell, or hold.

Will a stock split? How to tell if a stock is likely to split — the plain-English guide to the signals behind this score.
Stock split ratios explained — every ratio family (forward, reverse, fractional dividend, spin-off).
Stocks most likely to split — the live ranking this methodology powers.

Looking for a specific company? Browse the company directory and open the "Stock split history" tile on any ticker hub.